Kepler Documentation

Run Multiple Coding Agents in Parallel on Windows with Kepler

Last updated: November 2026

Kepler by GitKraken installs natively on Windows, so you can run Claude Code, Codex and other coding agents in parallel without a Mac or WSL. Each task gets its own worktree, a private Git working copy, so agents that run at the same time never change each other’s files.

Requirements and limits

  • Scope: Installing Kepler on Windows, installing coding agents there, and running several agents at once
  • Prerequisite: Windows 10+, on 64-bit or ARM64, and a GitKraken account to sign in to Kepler
  • Prerequisite: Your own account for each coding agent you use, such as a Claude or ChatGPT plan
  • Limits: Auggie, Pi and OpenCode need Node.js on Windows. OpenCode installs only through npm on Windows
  • Limits: A worktree separates files only. Agents still share ports, databases and other local services. See What a worktree does not separate

Quick Start

  1. Go to gitkraken.com/kepler/download, download the Windows installer for your computer (64-bit or ARM64), and run it.
  2. Launch Kepler and click Sign in with GitKraken.
  3. In Choose your coding agents, switch on Claude Code and Codex. To install an agent you do not have yet, click its tile.
  4. In Bring in your repos, check the repositories you work in, then click the button that adds them.
  5. Click New task or press Ctrl+N, write a prompt, attach a repository with Add repo or folder, and click Start. Kepler creates a worktree for the task and starts the agent in it.
  6. Start a second task the same way, and pick another agent in the New task window. Both agents now work at the same time, each in its own worktree.

How to install Kepler on Windows

Kepler has installers for 2 Windows architectures:

Installer For
64-bit Windows 10+ on an Intel or AMD processor
ARM64 Windows 10+ on an ARM processor
  1. Go to gitkraken.com/kepler/download and pick the Windows installer for your processor.
  2. Run the installer.
  3. Launch Kepler and sign in. A guided start then helps you set up agents and repositories.

Kepler runs directly on Windows. You do not need WSL, a virtual machine or a Mac to use it. See Getting Started with Kepler for the other platforms.


Kepler on Windows with 2 tasks running on the acme-web repo. The Tasks list shows Add product search and category filters and Apply discount codes to cart total under Running. Side by side, a Claude Code session with Opus 5.5 writes a reply and a Codex session with 6.1 Sol runs npm test. A teal ring marks the agent and model label in each panel: Default · Opus 5.5 under the Claude Code session and 6.1 Sol Medium under the Codex session. The Windows minimize, maximize and close buttons are at the top right.
Claude Code (Opus 5.5) and Codex (6.1 Sol) running side by side in Kepler on Windows, each in its own task.

How to install coding agents on Windows

Kepler supports 9 coding agents on Windows. When an agent is missing, open Settings → Agents, click Install in its section, and pick an install method. Kepler runs the install command for you and shows its output. After the install, Kepler offers to sign you in to the agent.

Agent Install methods on Windows Note for Windows
Claude Code Native installer, winget, npm (global) Kepler marks npm Deprecated by Anthropic
Codex Standalone installer, npm (global) None
GitHub Copilot winget, npm (global) None
Cursor Native installer None
Auggie npm (global) Needs Node.js
OpenCode npm (global) Needs Node.js. npm is the only Windows method
Grok Build Native installer None
Pi Native installer, npm (global) Needs Node.js for both methods. Runs in Terminal only
Antigravity Native installer Google’s agent

Kepler offers a method only when the tool it runs, such as winget or npm, is installed. When an agent needs Node.js and it is missing, Kepler says so and offers Get Node.js. Install Node.js, then click Rescan.

If you installed an agent yourself and Kepler did not find it, open the agent’s Set up dialog, enter the program’s full path under Already installed?, and click Use this. See Agent Integrations.


The Set up Claude Code dialog in Kepler on Windows. Under Install it lists Native installer, winget, and npm (global), marked Deprecated by Anthropic. Under Already installed? is a field for the full path to Claude Code with Browse and Use this buttons.
The Windows install routes for Claude Code.

How to run agents in parallel

In Kepler, the unit of work is a task. Each task holds its own sessions, and each repository you attach to it gets its own worktree. To run several agents at once, start one task per piece of work:

  1. Click New task or press Ctrl+N.
  2. Write the prompt, and attach the repository with Add repo or folder.
  3. Under Agent, pick the agent for this task, such as Claude Code.
  4. Click Start.
  5. Repeat for the next piece of work, with the same agent or another one, such as Codex.

You can attach the same repository to several tasks. Each task gets its own worktree on its own branch, so 2 agents can change the same repository at the same time.

To follow every agent at once, use the Dashboard. Kepler notifies you when an agent needs you, for example to answer a permission request. See Agent Sessions.


The New task button at the left end of the Kepler top bar, a single button with a plus sign, after the application menu, the window switcher and the Back and Forward buttons.
The New task button starts each parallel task.

How a worktree per task keeps agents apart

A worktree is a private Git working copy for one task. Agents make their changes there, so nothing they do touches the checkout you work in, and agents in different tasks never change each other’s files. This works the same on native Windows as on macOS and Linux.

Isolated worktree is the default for new tasks. You can change the default in Settings → Tasks → Where new tasks make changes, and change it for each repository in each task. See Tasks and Resources.


The Resources screen on its Repos & folders tab, adding from your repos and folders, with Open folder… and Clone…, the search field and the repositories folder /home/alex/kepler/repositories with Scan and Change. The acme-web row is selected, and under it a confirm line offers New worktree from main, reads Creates a worktree on a new branch from main, and has Cancel and Create worktree.
Adding a repository to a task creates a new worktree for it.

A new worktree has none of your Git-ignored files, such as .env or node_modules. To set up each new worktree automatically, see Set Up Every New Agent Worktree Automatically.

What a worktree does not separate

A worktree gives each task its own copy of the files, and nothing else. Agents in different tasks still share:

  • Ports. 2 worktrees that start the same dev server on the same port collide.
  • Databases and other local services. Agents that point at the same database read and write the same data.
  • Caches and files outside the repository, such as a global package cache.

To give each worktree its own port or database, use a repository command that runs when Kepler creates the worktree. See Set Up Every New Agent Worktree Automatically.


When to use WSL 2 instead

Run agents natively on Windows when your project builds and runs on Windows. Use a WSL 2 distro when your project needs a Linux toolchain, such as build scripts written for bash.

In Kepler, a WSL 2 distro is a remote environment. Your agents, worktrees and terminals run inside the distro, while you work from the Kepler window on Windows. It needs no credentials and no extra setup:

  1. Open Settings → Remote environments.
  2. Find your distro under Found on this machine, and click Add.
  3. Connect to it from the environment chip in the title bar.

Kepler lists WSL version 2 distros only. Install each agent inside the distro, because agents in a remote environment run there. See Remote Environments.


Common questions

Does Kepler work on Windows, or is it Mac only?

Kepler works on Windows. It has a native Windows installer for Windows 10+, on 64-bit and ARM64. It also runs on macOS and Linux.

Can I run Claude Code and Codex at the same time on Windows?

Yes. Start one task for each, and pick Claude Code for one and Codex for the other. Each task gets its own worktree, so the 2 agents never change each other’s files.

Do I need WSL to run coding agents on Windows?

No. All 9 agents that Kepler supports install and run on native Windows. Use WSL 2 only when your project needs a Linux toolchain.

How do I stop parallel agents from overwriting each other’s work?

Leave Isolated worktree on, which is the default. Each task then works in its own worktree on its own branch. When the work is done, review and merge each branch on its own. See Review Changes.

Why does my second agent’s dev server fail to start?

When 2 worktrees start a dev server on the same port, the second server can fail to start. A worktree separates files only. Give each worktree its own port with a repository command. See Set Up Every New Agent Worktree Automatically.


Related pages

Have feedback about this article? Did we miss something? Let us know!
On this page